Non-Contact Heartbeat Detection Based on Ballistocardiogram Using UNet and Bidirectional Long Short-Term Memory

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Non-invasive sensing technologies are crucial for home-care applications.
• Ballistocardiogram (BCG) signals offer a non-invasive method for physiological monitoring.
• Accurate heartbeat detection from BCG is essential for cardiovascular disease (CVD) risk prediction and sleep staging.

Purpose of the Study:

• To propose an effective deep learning model for automatic heartbeat detection from BCG signals.
• To address challenges in BCG-aided heartbeat detection, particularly in low signal-to-noise ratio (SNR) environments.
• To develop a reliable solution for long-term heart rate monitoring.

Main Methods:

• A deep learning model combining UNet and bidirectional long short-term memory (Bi-LSTM) was developed.
• The model was trained and validated using BCG recordings from 43 subjects.
• Performance was evaluated by comparing detected heartbeat intervals with R-R intervals from ECG, considering various postures and signal qualities.

Main Results:

• The proposed UNet-BiLSTM model achieved promising accuracy in detecting heartbeat intervals across different postures and signal qualities.
• The model demonstrated superior performance compared to state-of-the-art methods, with low mean absolute error and mean relative error.
• Numerical results confirmed the model's robustness against noise and perturbations like respiratory effort and artifact motion.

Conclusions:

• The developed UNet-BiLSTM model provides an effective and reliable solution for automatic heartbeat detection from BCG signals.
• The model is robust to noise and artifacts, making it suitable for real-world home-care applications.
• This technology facilitates accurate long-term heart rate monitoring, aiding in cardiovascular health management.